[Ffmpeg-devel-irc] ffmpeg-devel.log.20130911

[burek]

[01:25] <cone-545> ffmpeg.git 03Vignesh Venkatasubramanian 07master:ae12d655384a: lavcodec: Adding support for End Trimming in Opus encoder
[05:04] <cone-859> ffmpeg.git 03Michael Niedermayer 07master:7bc9fb8cde2b: avcodec/utils: add 16bit planar YUV formats to avcodec_align_dimensions2()
[05:04] <cone-859> ffmpeg.git 03Michael Niedermayer 07master:64b6279d1414: avcodec/mjpegdec: Support >8bpc color mjpeg
[11:37] <j-b> 'morning
[11:38] <Compn> so early
[11:39] <Compn> why'd you wake me up j-b ? :P
[11:41] <av500> j-b: im in paris tonight, can you deliver trocadero to my hotel? 
[11:56] <j-b> av500: no. but beers at my place
[11:56] <av500> wo ist das?
[12:07] <j-b> myplace.com
[12:08] <av500> ill be there
[12:08] <Daemon404> no habbo hotel?
[12:09] <av500> i haz hotel by the lake
[12:09] <av500> I hope its not a toxic waste dump
[12:11] <Compn> j-b : does voo have enough space to host some dcp samples ?
[12:12] <Daemon404> av500, it's paris...
[12:12] <Daemon404> everywhere is some form of toxic
[12:12] <Daemon404> due ot cigarettes
[12:12] <Compn> i think buxiness mentioned having some he had to upload...
[12:18] <Compn> ehe
[12:18] <Compn> ubitux : people want to do chromakey/ greenscreen videos with ffmpeg (e.g. person in front of green screen replaces green with video)
[12:19] <ubitux> why are you telling me this?
[12:19] <ubitux> don't we have something for this?
[12:19] <ubitux> i think we discussed that a while ago
[12:20] <Compn> i couldnt remember
[12:21] <Compn> i guess i should be asking saste ? 
[12:23] <saste> Compn, not sure it is possible, not without coding
[12:23] <saste> first we need some code to detect pixels in a given chroma interval, and create a mask from that
[12:23] <saste> then you can use overlay
[12:24] <saste> in theory geq could be used for the first stage, but i'm not sure it's so easy
[12:24] <durandal_1707> for first stage of what?
[12:25] <saste> durandal_1707, check which pixels are in a given chromatic interval or region
[12:26] <saste> HSV could be used (which reminds me of the fishdetect filter)
[12:27] <durandal_1707> and you need that for?
[12:28] <saste> durandal_1707, then you extract a mask, which corresponds to the region in the image corresponding to the given "background" color
[12:29] <saste> now you can use the mask to do an overlay, so you create a custom background which is overlaid on the background region of the original video
[12:30] <durandal_1707> i did it already in colorkeymask filter
[12:30] <ubitux> ah! see!
[12:30] <ubitux> i knew we discussed this and could do something
[12:31] <ubitux> durandal_1707: you didn't push it yet?
[12:31] <durandal_1707> its simple and crude, little better than internal avs filter 
[12:31] <cone-744> ffmpeg.git 03Alexandra Khirnova 07master:f369b9356c46: avformat: Use av_reallocp_array() where suitable
[12:31] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:cbe47b1e8452: Merge commit 'f369b9356c4606cd4d713d60f7db5de119d901fa'
[12:31] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:32805f852116: avformat/matroskadec: ebml_parse_elem() should not deallocate priorly existing lists on failure
[12:31] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:c9367c07091b: avformat/matroskaenc: functions that add entries should not destroy the whole list on failure
[12:31] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:5ca39a0b899a: avformat/mxfdec: functions that add entries should not destroy the whole list on failure
[12:31] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:8c9d6ee4f2ad: avformat/utils: functions that add entries should not destroy the whole list on failure
[12:32] <ubitux> durandal_1707: how dare you insult avs
[12:32] <durandal_1707> well there appears to be more poverful external avs filters
[12:32] <durandal_1707> but i doubt you will find what is used by pro
[12:32] <ubitux> ah i'm relieved then
[12:33] <durandal_1707> but we are really missing one such filter....
[12:34] <ubitux> yep
[12:35] <durandal_1707> i couldn't find how to correctly handle borders/transition
[12:35] <durandal_1707> you can read wikipedia entry for start..
[12:36] <durandal_1707> and find/make high-quality video with blue screen as background...
[12:40] <cone-744> ffmpeg.git 03Diego Biurrun 07master:c58f4069e1d5: fate.sh: Run git-clone quietly
[12:41] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:24da9eb25d63: Merge commit 'c58f4069e1d5b5804c669b691510e1b8fabb67fc'
[12:43] <ubitux> any idea what motivated the quite clone?
[12:43] <ubitux> i liked having the status..
[12:49] <michaelni> ubitux, no idea, and no objections to revert as far as iam concerned
[12:49] <ubitux> well it doesn't really matter
[12:49] <cone-744> ffmpeg.git 03Diego Biurrun 07master:ec17d1aa8b83: doc: Drop VDPAU from list of supported codecs
[12:50] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:c45e9097d839: Merge commit 'ec17d1aa8b833d49d8fc479e44e9e22858aa486c'
[12:50] <ubitux> i just like to see in my screen when there is an update of source files between instances run
[12:51] <Daemon404> why sit and watch automated fate runs?
[12:53] <ubitux> i happened to watch the status sometimes
[13:02] <cone-744> ffmpeg.git 03Diego Biurrun 07master:a0b901a348e4: Drop pointless directory name prefixes from #includes in the current dir
[13:02] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:803445e02c77: Merge remote-tracking branch 'qatar/master'
[13:13] <cone-744> ffmpeg.git 03Paul B Mahol 07master:00e1bf8a587e: avformat/flic: check for memory allocation failures
[13:13] <cone-744> ffmpeg.git 03Paul B Mahol 07master:60caa3d8bc11: avformat/sierravmd: check for memory allocation failures
[13:13] <cone-744> ffmpeg.git 03Paul B Mahol 07master:f2e614e7e9cb: avformat/ffmetadec: return meaningful error codes
[13:13] <cone-744> ffmpeg.git 03Paul B Mahol 07master:848ad8db2208: avformat/libnut: check avformat_new_stream() return value
[13:13] <cone-744> ffmpeg.git 03Paul B Mahol 07master:9e505485f335: avformat/smacker: check avformat_new_stream() return value
[13:29] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:a6ca08f1af31: avformat/gxfenc: Check and propagate return code from gxf_write_map_packet()
[13:29] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:2832a82e17e9: avformat/gxfenc: Clear nb_fields on array deallocation
[13:29] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:2175f7689409: avformat/asfenc: use av_freep() for extra saftey
[13:29] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:b5b34c198373: avformat/matroskaenc: use av_freep() to avoid leaving stale pointers
[13:49] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:e5290823697d: avformat/matroskadec: use av_realloc_array()
[13:49] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:9bca8e534150: avformat/matroskaenc: use av_realloc_array()
[13:49] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:c5714097eb4c: avformat/mxfdec: use av_realloc_array()
[13:49] <cone-744> ffmpeg.git 03Michael Niedermayer 07master:2a1c24a267e9: avformat/utils: use av_realloc_array()
[13:52] <michaelni> ubitux, mateo`, does one of you have time to look at todays mov.c changes ? i suspect some of it is not correct/ideal
[13:53] <ubitux> i'm not sure i'll have time today
[13:54] <michaelni> ubitux, ok, mateo` ?
[13:54] <michaelni> also doesnt need to be today ...
[13:58] <michaelni> if noone has time then ill look at it
[13:58] Action: michaelni hopes he doesnt forget
[14:01] <ubitux> michaelni: are you refering to the realloc thing?
[14:01] <michaelni> ubitux, yes
[14:02] <ubitux> you want to know if the empty list won't make everything explode?
[14:02] <michaelni> not only
[14:04] <michaelni> first question, is it a good idea? killing the lists is not very friendly if one wants to continue to use the demuxer
[14:04] <michaelni> and then of course yes, if its kept, all indexes that point into the arrays need to be reset too
[14:06] <ubitux> there are some cross references between the samples lists
[14:06] <ubitux> it should not be made inconsistent IMO
[14:07] <ubitux> if these realloc errors out, something nasty is going on, it might be relevant to error out
[14:07] <ubitux> but maybe you want to continue playback/decode anyway?
[14:10] <michaelni> yes
[14:12] <ubitux> i wonder why the checks above the alloc weren't removed
[14:12] <ubitux> since the reallocp_array should do them
[14:25] <ubitux> we should remove the ppa link to ffmpeg
[14:25] <ubitux> 0.10 is really too outdated
[14:32] <mateo`> where are the samples uploaded (k'!
[14:32] <mateo`> ?
[14:32] <mateo`> i'm looking for the sample related to https://trac.ffmpeg.org/ticket/2945
[14:36] <michaelni> ubitux, about 0.10, i suspect the old ubuntu versions need something soname compatible to that old version
[14:37] <michaelni> because there are heaps of binary packages linking to that soname
[14:37] <michaelni> newer ppas in addition are very needed though too
[14:38] <Compn> mateo` : on videolan server 
[14:40] <durandal_1707> mateo`: reply to ticker and demand link to file (even better: also close it with needs_more_info)
[14:41] <Compn> durandal_1707 : its here > http://streams.videolan.org/incoming/Test_file_ticket_2945_support_remux_m2v_into_mxf.MXF
[14:41] <Compn> pm me if you need login info
[14:42] <durandal_1707> cehoyos should reupload it and provide link in ticket
[14:46] <Compn> why? the incoming ftp is accessible by all devs ?
[14:46] <Compn> i can put it on samples repo if you want ?
[14:47] <Compn> durandal_1707 : do you and cehoyos have some secret war going on ? 
[14:47] <Compn> ehe
[14:48] <Daemon404> cehoyos drives me nuts too
[14:48] <Daemon404> i dont blame durandal_1707 
[14:49] <Compn> he is strict on the bugreports and requiring information to reproduce an issue
[14:49] <Compn> do you want a more neutral place to report bugs ?
[14:50] <Daemon404> no he's hypocritical and uses blanket responses
[14:50] <Compn> maybe we could ask carl to go easy on ffmpeg devels like durandal_1707 and Daemon404 :)
[14:50] <Compn> ehehe
[14:50] <Daemon404> and cloes valid bugs with proper info with generic blanket statements
[14:50] <Daemon404> and doesnt follow the patch contribution rules
[14:50] <Daemon404> despite harping on others to do so
[14:50] <Daemon404> learn to use git-fucking-format-patch
[14:50] <Daemon404> :V
[14:51] <Compn> lol
[14:51] <Compn> hey i dont use that either :P
[14:51] <Daemon404> you dont submit patches
[14:52] <Daemon404> (except to vlc, where remi trolls you)
[14:53] <Compn> i dont get why vlc needs two entries for the same codec :P
[15:02] <durandal_1707> you are all overreacting
[15:42] <durandal_1707> mdsh: /win3
[15:43] <durandal_1707> sorry, now i'm really forced to ask you
[15:43] <durandal_1707> mdsh: do you plan to send any patches to ffmpeg ml?
[15:49] <mdsh> durandal_1707: I plan to send more patches to the ml, yes. Why - what have I missed...
[15:51] <durandal_1707> regarding w3fdif filter
[15:58] <mdsh> durandal_1707: when I get the time I plan to make w3fdif do >8 bit and try to increase the speed (for instance, michealni's suggestion of working on 32x32 rectangles - allowing multi-threading)
[15:58] <mdsh> Unless anyone else does it first...
[16:03] <durandal_1707> why 32x32 rectangles are needed for mt?
[16:07] <mdsh> michael said in an email about 2 weeks ago that doing smaller rectangles should be better for CPU cache hits - and implementing that will make MT easy
[16:08] <Daemon404> i havent looked at the code, but does w3fdif not have dependences on previous pixels?
[16:08] <Daemon404> or does it only need pre-intrpolation values?
[16:08] <mdsh> no, each output pixel is just a convolution of original surrounding pixels
[16:09] <Daemon404> ah ok.
[16:10] <durandal_1707> hmm, that would apply for single case too, and there are other filters that could make use of it...
[16:12] <durandal_1707> if gain is > 10% i would do it right now for transpose filter...
[16:12] <durandal_1707> i'm afraid if i do it, to find out it helps 0.1 %
[16:12] <Daemon404> our transpose filter can be made a *lot* faster as it is
[16:12] <Daemon404> without mt
[16:13] <Daemon404> it's very slow
[16:13] <Daemon404> i had a patch to add vapoursynth's transpose ASM at one point
[16:14] <mdsh> it would be fabulous if someone could create some ASM for w3fdif too - beyond my skill level
[16:14] <durandal_1707> well that would need smaller rectangles?
[16:15] <Daemon404> mdsh, btw is w3fdif implementable as a general convolution
[16:15] <Daemon404> or is it special?
[16:15] <durandal_1707> Daemon404: how much is vs transpose fast?
[16:15] <Daemon404> it was more than 2x
[16:15] <durandal_1707> compared to lavfi one (without mt)
[16:15] <Daemon404> when i tried
[16:15] <Daemon404> maybe it was 3x, cant remember
[16:15] <Daemon404> it was nontrivial
[16:16] <durandal_1707> it is in internal?
[16:16] <Daemon404> the patch?
[16:16] <Daemon404> i lost it
[16:16] <Daemon404> the asm is public
[16:21] <durandal_1707> i found asm and its lgpl
[16:22] <Daemon404> yes
[16:37] <durandal_1707> what you wanted to do with (now lost) patch?
[16:39] <Daemon404> i was going to submit it
[16:39] <Daemon404> but i forgot
[16:43] <durandal_1707> and it was fully working?
[16:55] <Daemon404> durandal_1707, mostly. 
[16:57] <durandal_1707> what that means?
[16:58] <Daemon404> it had a few edge cases?
[16:58] <Daemon404> what does it matter?
[16:58] <Daemon404> the patch no longer exists
[16:58] <durandal_1707> :((
[16:58] <gnafu> "<@Daemon404> durandal_1707, mostly." <-- That's when they come out at night, right?
[17:00] <cone-744> ffmpeg.git 03Dylan Alex Simon 07master:38155b47fd56: avcodec: add const qualifier to avcodec_find_best_pix_fmt2 args
[17:03] <cone-744> ffmpeg.git 03Paul B Mahol 07master:99a283331c8e: avfilter/vf_traspose: move switch out of loop
[17:04] <cone-744> ffmpeg.git 03Paul B Mahol 07master:269845db86d9: avformat/nutdec: check for allocation failures
[17:05] <cone-744> ffmpeg.git 03Paul B Mahol 07master:a8e00cf926f3: avfilter: remove redundant use of AV_NE() macro
[17:05] <ubitux> durandal_1707: why dst is incremented but not src?
[17:05] <ubitux> not that it's a problem, but it's inconsistent :p
[17:15] <durandal_1707> libavformat/asfenc.c:872:18: warning: unsequenced modification and access to 'ret' [-Wunsequenced]
[17:20] <durandal_1707> ubitux: so what is 2.1 release's name?
[17:33] <ubitux> durandal_1707: quezac
[17:34] <ubitux> well, private french joke i believe
[17:34] <ubitux> forget this
[17:38] <mateo`> haha
[17:38] <durandal_1707> why they don't teach (y)asm in scool
[17:40] <mateo`> durandal_1707: they tends to prefer teaching web things instead ... the future you know :)
[19:01] <durandal_1707> ok i hacked something, but when i run it i only get bus error
[19:15] <durandal_1707> oh, got it, code is obviously ssse3 only
[19:20] <durandal_1707> or not :(
[19:29] <wm4> would it be possible to add the number of skipped samples to an AVFrame?
[19:33] <durandal_1707> hmm, it seems there is already some transpose code in lavc
[19:49] <durandal_1707> Daemon404: you sure that asm ever worked?
[19:58] <Daemon404> yes
[20:01] <durandal_1707> on 32bit?
[20:17] <durandal_1707> also code is little clusmsy, why it need 8 instruction to read 8 bytes....
[20:19] <durandal_1707> ah, it reads 8 bytes and not single one
[20:51] <wm4> does anyone know whether ingenientdec.c is actually supposed to handle this?: http://128.104.138.26/axis-cgi/mjpg/video.cgi
[20:52] <durandal_1707> maybe
[20:52] <durandal_1707> fetch it, do you see bunch MJPG in it?
[20:52] <wm4> no
[20:53] <wm4> it seems to be rather different
[20:53] <wm4> but who knows
[20:54] <wm4> hm I think it's just jpegs dumped as multi-part thing
[20:54] <wm4> libavcodec can actually decode the raw stream as jpg (I suppose it skips the header as garbage)
[20:55] <wm4> would probably be easy to write a libavformat demuxer for this
[20:57] <Daemon404> everyone and their grandma has a custom container with jpeg in it
[20:57] <Daemon404> :V
[21:08] <durandal_1707> s/grandma/dog
[21:10] <durandal_1707> how to move another half (like with movh)?
[21:37] <durandal_1707> simply using movhps with >= sse2 i get crash
[21:38] <durandal_1707> using just movh with mmx i do not get crash but result is half broken
[21:38] <durandal_1707> so it code does: movh [dstq], m0
[21:38] <durandal_1707> how to get other part of m0 ?
[22:26] <cone-744> ffmpeg.git 03James Almer 07master:7e4fe5162ab9: lavu/sha: Fully unroll the transform function loops
[22:26] <cone-744> ffmpeg.git 03James Almer 07master:bbcaf25d4d21: lavu/sha512: Fully unroll the transform function loops
[00:00] --- Thu Sep 12 2013